Threading and performance

For the QR decomposition, much of the time spent in MPO construction is in the sparse QR decomposition. This can often be sped up significantly by loading MKL.

try
  using MKL
catch
end

ITensorMPOConstruction can take advantage of multiple threads in a variety of ways.

1. ITensorMPS.add!(os::OpIDSum, ...) is thread-safe. This can be used to construct an OpIDSum in parallel.

2. During MPO construction, Threads.@threads is used primarily to iterate over the connected components (i.e. QN blocks) in parallel.

3. With alg="QR", sparse QR decompositions are performed for connected components. The threading here is controlled by BLAS.set_num_threads. Oversubscription is possible since this QR decomposition is itself called from within Threads.@threads.

Threading tips

• MPO construction is memory-bound, and the performance gain (or loss!) from multithreading is system and operator dependent.
• Change the number of threads used for garbage collection.
• If the MPO has lots of connected components (i.e. is sparse), use Julia threads.
• If the MPO has few connected components (i.e. is dense), try using BLAS threads.